4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/module.h>
19 #include <linux/slab.h>
21 #include <linux/namei.h>
22 #include <linux/quotaops.h>
23 #include <linux/pagemap.h>
24 #include <linux/dnotify.h>
25 #include <linux/smp_lock.h>
26 #include <linux/personality.h>
27 #include <linux/security.h>
28 #include <linux/mount.h>
29 #include <linux/audit.h>
30 #include <asm/namei.h>
31 #include <asm/uaccess.h>
33 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
35 /* [Feb-1997 T. Schoebel-Theuer]
36 * Fundamental changes in the pathname lookup mechanisms (namei)
37 * were necessary because of omirr. The reason is that omirr needs
38 * to know the _real_ pathname, not the user-supplied one, in case
39 * of symlinks (and also when transname replacements occur).
41 * The new code replaces the old recursive symlink resolution with
42 * an iterative one (in case of non-nested symlink chains). It does
43 * this with calls to <fs>_follow_link().
44 * As a side effect, dir_namei(), _namei() and follow_link() are now
45 * replaced with a single function lookup_dentry() that can handle all
46 * the special cases of the former code.
48 * With the new dcache, the pathname is stored at each inode, at least as
49 * long as the refcount of the inode is positive. As a side effect, the
50 * size of the dcache depends on the inode cache and thus is dynamic.
52 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
53 * resolution to correspond with current state of the code.
55 * Note that the symlink resolution is not *completely* iterative.
56 * There is still a significant amount of tail- and mid- recursion in
57 * the algorithm. Also, note that <fs>_readlink() is not used in
58 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
59 * may return different results than <fs>_follow_link(). Many virtual
60 * filesystems (including /proc) exhibit this behavior.
63 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
64 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
65 * and the name already exists in form of a symlink, try to create the new
66 * name indicated by the symlink. The old code always complained that the
67 * name already exists, due to not following the symlink even if its target
68 * is nonexistent. The new semantics affects also mknod() and link() when
69 * the name is a symlink pointing to a non-existant name.
71 * I don't know which semantics is the right one, since I have no access
72 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
73 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
74 * "old" one. Personally, I think the new semantics is much more logical.
75 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
76 * file does succeed in both HP-UX and SunOs, but not in Solaris
77 * and in the old Linux semantics.
80 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
81 * semantics. See the comments in "open_namei" and "do_link" below.
83 * [10-Sep-98 Alan Modra] Another symlink change.
86 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
87 * inside the path - always follow.
88 * in the last component in creation/removal/renaming - never follow.
89 * if LOOKUP_FOLLOW passed - follow.
90 * if the pathname has trailing slashes - follow.
91 * otherwise - don't follow.
92 * (applied in that order).
94 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
95 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
96 * During the 2.4 we need to fix the userland stuff depending on it -
97 * hopefully we will be able to get rid of that wart in 2.5. So far only
98 * XEmacs seems to be relying on it...
101 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
102 * implemented. Let's see if raised priority of ->s_vfs_rename_sem gives
103 * any extra contention...
106 /* In order to reduce some races, while at the same time doing additional
107 * checking and hopefully speeding things up, we copy filenames to the
108 * kernel data space before using them..
110 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
111 * PATH_MAX includes the nul terminator --RR.
113 static inline int do_getname(const char __user *filename, char *page)
116 unsigned long len = PATH_MAX;
118 if ((unsigned long) filename >= TASK_SIZE) {
119 if (!segment_eq(get_fs(), KERNEL_DS))
121 } else if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
122 len = TASK_SIZE - (unsigned long) filename;
124 retval = strncpy_from_user((char *)page, filename, len);
128 return -ENAMETOOLONG;
134 char * getname(const char __user * filename)
138 result = ERR_PTR(-ENOMEM);
141 int retval = do_getname(filename, tmp);
146 result = ERR_PTR(retval);
149 if (unlikely(current->audit_context) && !IS_ERR(result) && result)
150 audit_getname(result);
157 * is used to check for read/write/execute permissions on a file.
158 * We use "fsuid" for this, letting us set arbitrary permissions
159 * for filesystem access without changing the "normal" uids which
160 * are used for other things..
162 int vfs_permission(struct inode * inode, int mask)
164 umode_t mode = inode->i_mode;
166 if (mask & MAY_WRITE) {
168 * Nobody gets write access to a read-only fs.
170 if (IS_RDONLY(inode) &&
171 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
175 * Nobody gets write access to an immutable file.
177 if (IS_IMMUTABLE(inode))
181 if (current->fsuid == inode->i_uid)
183 else if (in_group_p(inode->i_gid))
187 * If the DACs are ok we don't need any capability check.
189 if (((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask))
193 * Read/write DACs are always overridable.
194 * Executable DACs are overridable if at least one exec bit is set.
196 if (!(mask & MAY_EXEC) ||
197 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
198 if (capable(CAP_DAC_OVERRIDE))
202 * Searching includes executable on directories, else just read.
204 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
205 if (capable(CAP_DAC_READ_SEARCH))
211 int permission(struct inode * inode,int mask, struct nameidata *nd)
216 /* Ordinary permission routines do not understand MAY_APPEND. */
217 submask = mask & ~MAY_APPEND;
219 if (inode->i_op && inode->i_op->permission)
220 retval = inode->i_op->permission(inode, submask, nd);
222 retval = vfs_permission(inode, submask);
226 return security_inode_permission(inode, mask, nd);
230 * get_write_access() gets write permission for a file.
231 * put_write_access() releases this write permission.
232 * This is used for regular files.
233 * We cannot support write (and maybe mmap read-write shared) accesses and
234 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
235 * can have the following values:
236 * 0: no writers, no VM_DENYWRITE mappings
237 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
238 * > 0: (i_writecount) users are writing to the file.
240 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
241 * except for the cases where we don't hold i_writecount yet. Then we need to
242 * use {get,deny}_write_access() - these functions check the sign and refuse
243 * to do the change if sign is wrong. Exclusion between them is provided by
244 * the inode->i_lock spinlock.
247 int get_write_access(struct inode * inode)
249 spin_lock(&inode->i_lock);
250 if (atomic_read(&inode->i_writecount) < 0) {
251 spin_unlock(&inode->i_lock);
254 atomic_inc(&inode->i_writecount);
255 spin_unlock(&inode->i_lock);
260 int deny_write_access(struct file * file)
262 struct inode *inode = file->f_dentry->d_inode;
264 spin_lock(&inode->i_lock);
265 if (atomic_read(&inode->i_writecount) > 0) {
266 spin_unlock(&inode->i_lock);
269 atomic_dec(&inode->i_writecount);
270 spin_unlock(&inode->i_lock);
275 void path_release(struct nameidata *nd)
282 * Internal lookup() using the new generic dcache.
285 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
287 struct dentry * dentry = __d_lookup(parent, name);
289 /* lockess __d_lookup may fail due to concurrent d_move()
290 * in some unrelated directory, so try with d_lookup
293 dentry = d_lookup(parent, name);
295 if (dentry && dentry->d_op && dentry->d_op->d_revalidate) {
296 if (!dentry->d_op->d_revalidate(dentry, nd) && !d_invalidate(dentry)) {
305 * Short-cut version of permission(), for calling by
306 * path_walk(), when dcache lock is held. Combines parts
307 * of permission() and vfs_permission(), and tests ONLY for
308 * MAY_EXEC permission.
310 * If appropriate, check DAC only. If not appropriate, or
311 * short-cut DAC fails, then call permission() to do more
312 * complete permission check.
314 static inline int exec_permission_lite(struct inode *inode,
315 struct nameidata *nd)
317 umode_t mode = inode->i_mode;
319 if ((inode->i_op && inode->i_op->permission))
322 if (current->fsuid == inode->i_uid)
324 else if (in_group_p(inode->i_gid))
330 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
333 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
338 return security_inode_permission(inode, MAY_EXEC, nd);
342 * This is called when everything else fails, and we actually have
343 * to go to the low-level filesystem to find out what we should do..
345 * We get the directory semaphore, and after getting that we also
346 * make sure that nobody added the entry to the dcache in the meantime..
349 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
351 struct dentry * result;
352 struct inode *dir = parent->d_inode;
356 * First re-do the cached lookup just in case it was created
357 * while we waited for the directory semaphore..
359 * FIXME! This could use version numbering or similar to
360 * avoid unnecessary cache lookups.
362 * The "dcache_lock" is purely to protect the RCU list walker
363 * from concurrent renames at this point (we mustn't get false
364 * negatives from the RCU list walk here, unlike the optimistic
367 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
369 result = d_lookup(parent, name);
371 struct dentry * dentry = d_alloc(parent, name);
372 result = ERR_PTR(-ENOMEM);
374 result = dir->i_op->lookup(dir, dentry, nd);
385 * Uhhuh! Nasty case: the cache was re-populated while
386 * we waited on the semaphore. Need to revalidate.
389 if (result->d_op && result->d_op->d_revalidate) {
390 if (!result->d_op->d_revalidate(result, nd) && !d_invalidate(result)) {
392 result = ERR_PTR(-ENOENT);
398 inline void nd_set_link(struct nameidata *nd, char *path)
400 nd->saved_names[current->link_count] = path;
403 inline char *nd_get_link(struct nameidata *nd)
405 return nd->saved_names[current->link_count];
408 EXPORT_SYMBOL(nd_set_link);
409 EXPORT_SYMBOL(nd_get_link);
411 static inline int __vfs_follow_link(struct nameidata *, const char *);
414 * This limits recursive symlink follows to 8, while
415 * limiting consecutive symlinks to 40.
417 * Without that kind of total limit, nasty chains of consecutive
418 * symlinks can cause almost arbitrarily long lookups.
420 static inline int do_follow_link(struct dentry *dentry, struct nameidata *nd)
423 if (current->link_count >= MAX_NESTED_LINKS)
425 if (current->total_link_count >= 40)
428 err = security_inode_follow_link(dentry, nd);
431 current->link_count++;
432 current->total_link_count++;
433 touch_atime(nd->mnt, dentry);
434 nd_set_link(nd, NULL);
435 err = dentry->d_inode->i_op->follow_link(dentry, nd);
437 char *s = nd_get_link(nd);
439 err = __vfs_follow_link(nd, s);
440 if (dentry->d_inode->i_op->put_link)
441 dentry->d_inode->i_op->put_link(dentry, nd);
443 current->link_count--;
450 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
452 struct vfsmount *parent;
453 struct dentry *mountpoint;
454 spin_lock(&vfsmount_lock);
455 parent=(*mnt)->mnt_parent;
456 if (parent == *mnt) {
457 spin_unlock(&vfsmount_lock);
461 mountpoint=dget((*mnt)->mnt_mountpoint);
462 spin_unlock(&vfsmount_lock);
464 *dentry = mountpoint;
470 /* no need for dcache_lock, as serialization is taken care in
473 static int follow_mount(struct vfsmount **mnt, struct dentry **dentry)
476 while (d_mountpoint(*dentry)) {
477 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
483 *dentry = dget(mounted->mnt_root);
489 /* no need for dcache_lock, as serialization is taken care in
492 static inline int __follow_down(struct vfsmount **mnt, struct dentry **dentry)
494 struct vfsmount *mounted;
496 mounted = lookup_mnt(*mnt, *dentry);
501 *dentry = dget(mounted->mnt_root);
507 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
509 return __follow_down(mnt,dentry);
512 static inline void follow_dotdot(struct vfsmount **mnt, struct dentry **dentry)
515 struct vfsmount *parent;
516 struct dentry *old = *dentry;
518 read_lock(¤t->fs->lock);
519 if (*dentry == current->fs->root &&
520 *mnt == current->fs->rootmnt) {
521 read_unlock(¤t->fs->lock);
524 read_unlock(¤t->fs->lock);
525 spin_lock(&dcache_lock);
526 if (*dentry != (*mnt)->mnt_root) {
527 *dentry = dget((*dentry)->d_parent);
528 spin_unlock(&dcache_lock);
532 spin_unlock(&dcache_lock);
533 spin_lock(&vfsmount_lock);
534 parent = (*mnt)->mnt_parent;
535 if (parent == *mnt) {
536 spin_unlock(&vfsmount_lock);
540 *dentry = dget((*mnt)->mnt_mountpoint);
541 spin_unlock(&vfsmount_lock);
546 follow_mount(mnt, dentry);
550 struct vfsmount *mnt;
551 struct dentry *dentry;
555 * It's more convoluted than I'd like it to be, but... it's still fairly
556 * small and for now I'd prefer to have fast path as straight as possible.
557 * It _is_ time-critical.
559 static int do_lookup(struct nameidata *nd, struct qstr *name,
562 struct vfsmount *mnt = nd->mnt;
563 struct dentry *dentry = __d_lookup(nd->dentry, name);
567 if (dentry->d_op && dentry->d_op->d_revalidate)
568 goto need_revalidate;
571 path->dentry = dentry;
575 dentry = real_lookup(nd->dentry, name, nd);
581 if (dentry->d_op->d_revalidate(dentry, nd))
583 if (d_invalidate(dentry))
589 return PTR_ERR(dentry);
595 * This is the basic name resolution function, turning a pathname
596 * into the final dentry.
598 * We expect 'base' to be positive and a directory.
600 int fastcall link_path_walk(const char * name, struct nameidata *nd)
605 unsigned int lookup_flags = nd->flags;
607 atomic = (lookup_flags & LOOKUP_ATOMIC);
614 inode = nd->dentry->d_inode;
615 if (current->link_count)
616 lookup_flags = LOOKUP_FOLLOW;
618 /* At this point we know we have a real path component. */
624 err = exec_permission_lite(inode, nd);
625 if (err == -EAGAIN) {
626 err = permission(inode, MAY_EXEC, nd);
632 c = *(const unsigned char *)name;
634 hash = init_name_hash();
637 hash = partial_name_hash(c, hash);
638 c = *(const unsigned char *)name;
639 } while (c && (c != '/'));
640 this.len = name - (const char *) this.name;
641 this.hash = end_name_hash(hash);
643 /* remove trailing slashes? */
646 while (*++name == '/');
648 goto last_with_slashes;
651 * "." and ".." are special - ".." especially so because it has
652 * to be able to know about the current root directory and
653 * parent relationships.
655 if (this.name[0] == '.') switch (this.len) {
659 if (this.name[1] != '.')
661 follow_dotdot(&nd->mnt, &nd->dentry);
662 inode = nd->dentry->d_inode;
668 * See if the low-level filesystem might want
669 * to use its own hash..
671 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
672 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
676 err = -EWOULDBLOCKIO;
679 nd->flags |= LOOKUP_CONTINUE;
680 /* This does the actual lookups.. */
681 err = do_lookup(nd, &this, &next);
684 /* Check mountpoints.. */
685 follow_mount(&next.mnt, &next.dentry);
688 inode = next.dentry->d_inode;
695 if (inode->i_op->follow_link) {
697 err = do_follow_link(next.dentry, nd);
703 inode = nd->dentry->d_inode;
712 nd->dentry = next.dentry;
715 if (!inode->i_op->lookup)
718 /* here ends the main loop */
721 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
723 nd->flags &= ~LOOKUP_CONTINUE;
724 if (lookup_flags & LOOKUP_PARENT)
726 if (this.name[0] == '.') switch (this.len) {
730 if (this.name[1] != '.')
732 follow_dotdot(&nd->mnt, &nd->dentry);
733 inode = nd->dentry->d_inode;
738 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
739 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
743 err = -EWOULDBLOCKIO;
746 err = do_lookup(nd, &this, &next);
749 follow_mount(&next.mnt, &next.dentry);
750 inode = next.dentry->d_inode;
751 if ((lookup_flags & LOOKUP_FOLLOW)
752 && inode && inode->i_op && inode->i_op->follow_link) {
754 err = do_follow_link(next.dentry, nd);
759 inode = nd->dentry->d_inode;
763 nd->dentry = next.dentry;
768 if (lookup_flags & LOOKUP_DIRECTORY) {
770 if (!inode->i_op || !inode->i_op->lookup)
776 nd->last_type = LAST_NORM;
777 if (this.name[0] != '.')
780 nd->last_type = LAST_DOT;
781 else if (this.len == 2 && this.name[1] == '.')
782 nd->last_type = LAST_DOTDOT;
787 * We bypassed the ordinary revalidation routines.
788 * We may need to check the cached dentry for staleness.
790 if (nd->dentry && nd->dentry->d_sb &&
791 (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
793 /* Note: we do not d_invalidate() */
794 if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd))
808 int fastcall path_walk(const char * name, struct nameidata *nd)
810 current->total_link_count = 0;
811 return link_path_walk(name, nd);
815 /* returns 1 if everything is done */
816 static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
818 if (path_walk(name, nd))
819 return 0; /* something went wrong... */
821 if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) {
822 struct nameidata nd_root;
824 * NAME was not found in alternate root or it's a directory. Try to find
825 * it in the normal root:
827 nd_root.last_type = LAST_ROOT;
828 nd_root.flags = nd->flags;
829 memcpy(&nd_root.intent, &nd->intent, sizeof(nd_root.intent));
830 read_lock(¤t->fs->lock);
831 nd_root.mnt = mntget(current->fs->rootmnt);
832 nd_root.dentry = dget(current->fs->root);
833 read_unlock(¤t->fs->lock);
834 if (path_walk(name, &nd_root))
836 if (nd_root.dentry->d_inode) {
838 nd->dentry = nd_root.dentry;
839 nd->mnt = nd_root.mnt;
840 nd->last = nd_root.last;
843 path_release(&nd_root);
848 void set_fs_altroot(void)
850 char *emul = __emul_prefix();
852 struct vfsmount *mnt = NULL, *oldmnt;
853 struct dentry *dentry = NULL, *olddentry;
858 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
864 write_lock(¤t->fs->lock);
865 oldmnt = current->fs->altrootmnt;
866 olddentry = current->fs->altroot;
867 current->fs->altrootmnt = mnt;
868 current->fs->altroot = dentry;
869 write_unlock(¤t->fs->lock);
878 walk_init_root(const char *name, struct nameidata *nd)
880 read_lock(¤t->fs->lock);
881 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
882 nd->mnt = mntget(current->fs->altrootmnt);
883 nd->dentry = dget(current->fs->altroot);
884 read_unlock(¤t->fs->lock);
885 if (__emul_lookup_dentry(name,nd))
887 read_lock(¤t->fs->lock);
889 nd->mnt = mntget(current->fs->rootmnt);
890 nd->dentry = dget(current->fs->root);
891 read_unlock(¤t->fs->lock);
895 int fastcall path_lookup(const char *name, unsigned int flags, struct nameidata *nd)
899 nd->last_type = LAST_ROOT; /* if there are only slashes... */
902 read_lock(¤t->fs->lock);
904 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
905 nd->mnt = mntget(current->fs->altrootmnt);
906 nd->dentry = dget(current->fs->altroot);
907 read_unlock(¤t->fs->lock);
908 if (__emul_lookup_dentry(name,nd))
910 read_lock(¤t->fs->lock);
912 nd->mnt = mntget(current->fs->rootmnt);
913 nd->dentry = dget(current->fs->root);
916 nd->mnt = mntget(current->fs->pwdmnt);
917 nd->dentry = dget(current->fs->pwd);
919 read_unlock(¤t->fs->lock);
920 current->total_link_count = 0;
921 retval = link_path_walk(name, nd);
922 if (unlikely(current->audit_context
923 && nd && nd->dentry && nd->dentry->d_inode))
925 nd->dentry->d_inode->i_ino,
926 nd->dentry->d_inode->i_rdev);
931 * Restricted form of lookup. Doesn't follow links, single-component only,
932 * needs parent already locked. Doesn't follow mounts.
935 static struct dentry * __lookup_hash(struct qstr *name, struct dentry * base, struct nameidata *nd)
937 struct dentry * dentry;
941 inode = base->d_inode;
942 err = permission(inode, MAY_EXEC, nd);
943 dentry = ERR_PTR(err);
948 * See if the low-level filesystem might want
949 * to use its own hash..
951 if (base->d_op && base->d_op->d_hash) {
952 err = base->d_op->d_hash(base, name);
953 dentry = ERR_PTR(err);
958 dentry = cached_lookup(base, name, nd);
960 struct dentry *new = d_alloc(base, name);
961 dentry = ERR_PTR(-ENOMEM);
964 dentry = inode->i_op->lookup(inode, new, nd);
974 struct dentry * lookup_hash(struct qstr *name, struct dentry * base)
976 return __lookup_hash(name, base, NULL);
980 struct dentry * lookup_one_len(const char * name, struct dentry * base, int len)
991 hash = init_name_hash();
993 c = *(const unsigned char *)name++;
994 if (c == '/' || c == '\0')
996 hash = partial_name_hash(c, hash);
998 this.hash = end_name_hash(hash);
1000 return lookup_hash(&this, base);
1002 return ERR_PTR(-EACCES);
1008 * is used by most simple commands to get the inode of a specified name.
1009 * Open, link etc use their own routines, but this is enough for things
1012 * namei exists in two versions: namei/lnamei. The only difference is
1013 * that namei follows links, while lnamei does not.
1016 int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1018 char *tmp = getname(name);
1019 int err = PTR_ERR(tmp);
1022 err = path_lookup(tmp, flags, nd);
1029 * It's inline, so penalty for filesystems that don't use sticky bit is
1032 static inline int check_sticky(struct inode *dir, struct inode *inode)
1034 if (!(dir->i_mode & S_ISVTX))
1036 if (inode->i_uid == current->fsuid)
1038 if (dir->i_uid == current->fsuid)
1040 return !capable(CAP_FOWNER);
1044 * Check whether we can remove a link victim from directory dir, check
1045 * whether the type of victim is right.
1046 * 1. We can't do it if dir is read-only (done in permission())
1047 * 2. We should have write and exec permissions on dir
1048 * 3. We can't remove anything from append-only dir
1049 * 4. We can't do anything with immutable dir (done in permission())
1050 * 5. If the sticky bit on dir is set we should either
1051 * a. be owner of dir, or
1052 * b. be owner of victim, or
1053 * c. have CAP_FOWNER capability
1054 * 6. If the victim is append-only or immutable we can't do antyhing with
1055 * links pointing to it.
1056 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1057 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1058 * 9. We can't remove a root or mountpoint.
1059 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1060 * nfs_async_unlink().
1062 static inline int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1065 if (!victim->d_inode)
1067 if (victim->d_parent->d_inode != dir)
1070 error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
1075 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1076 IS_IMMUTABLE(victim->d_inode))
1079 if (!S_ISDIR(victim->d_inode->i_mode))
1081 if (IS_ROOT(victim))
1083 } else if (S_ISDIR(victim->d_inode->i_mode))
1085 if (IS_DEADDIR(dir))
1087 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1092 /* Check whether we can create an object with dentry child in directory
1094 * 1. We can't do it if child already exists (open has special treatment for
1095 * this case, but since we are inlined it's OK)
1096 * 2. We can't do it if dir is read-only (done in permission())
1097 * 3. We should have write and exec permissions on dir
1098 * 4. We can't do it if dir is immutable (done in permission())
1100 static inline int may_create(struct inode *dir, struct dentry *child,
1101 struct nameidata *nd)
1105 if (IS_DEADDIR(dir))
1107 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1111 * Special case: O_CREAT|O_EXCL implies O_NOFOLLOW for security
1114 * O_DIRECTORY translates into forcing a directory lookup.
1116 static inline int lookup_flags(unsigned int f)
1118 unsigned long retval = LOOKUP_FOLLOW;
1121 retval &= ~LOOKUP_FOLLOW;
1123 if ((f & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1124 retval &= ~LOOKUP_FOLLOW;
1126 if (f & O_DIRECTORY)
1127 retval |= LOOKUP_DIRECTORY;
1128 if (f & O_ATOMICLOOKUP)
1129 retval |= LOOKUP_ATOMIC;
1135 * p1 and p2 should be directories on the same fs.
1137 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1142 down(&p1->d_inode->i_sem);
1146 down(&p1->d_inode->i_sb->s_vfs_rename_sem);
1148 for (p = p1; p->d_parent != p; p = p->d_parent) {
1149 if (p->d_parent == p2) {
1150 down(&p2->d_inode->i_sem);
1151 down(&p1->d_inode->i_sem);
1156 for (p = p2; p->d_parent != p; p = p->d_parent) {
1157 if (p->d_parent == p1) {
1158 down(&p1->d_inode->i_sem);
1159 down(&p2->d_inode->i_sem);
1164 down(&p1->d_inode->i_sem);
1165 down(&p2->d_inode->i_sem);
1169 void unlock_rename(struct dentry *p1, struct dentry *p2)
1171 up(&p1->d_inode->i_sem);
1173 up(&p2->d_inode->i_sem);
1174 up(&p1->d_inode->i_sb->s_vfs_rename_sem);
1178 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1179 struct nameidata *nd)
1181 int error = may_create(dir, dentry, nd);
1186 if (!dir->i_op || !dir->i_op->create)
1187 return -EACCES; /* shouldn't it be ENOSYS? */
1190 error = security_inode_create(dir, dentry, mode);
1194 error = dir->i_op->create(dir, dentry, mode, nd);
1196 inode_dir_notify(dir, DN_CREATE);
1197 security_inode_post_create(dir, dentry, mode);
1202 int may_open(struct nameidata *nd, int acc_mode, int flag)
1204 struct dentry *dentry = nd->dentry;
1205 struct inode *inode = dentry->d_inode;
1211 if (S_ISLNK(inode->i_mode))
1214 if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1217 error = permission(inode, acc_mode, nd);
1222 * FIFO's, sockets and device files are special: they don't
1223 * actually live on the filesystem itself, and as such you
1224 * can write to them even if the filesystem is read-only.
1226 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1228 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1229 if (nd->mnt->mnt_flags & MNT_NODEV)
1233 } else if (IS_RDONLY(inode) && (flag & FMODE_WRITE))
1236 * An append-only file must be opened in append mode for writing.
1238 if (IS_APPEND(inode)) {
1239 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1246 * Ensure there are no outstanding leases on the file.
1248 error = break_lease(inode, flag);
1252 if (flag & O_TRUNC) {
1253 error = get_write_access(inode);
1258 * Refuse to truncate files with mandatory locks held on them.
1260 error = locks_verify_locked(inode);
1264 error = do_truncate(dentry, 0);
1266 put_write_access(inode);
1270 if (flag & FMODE_WRITE)
1279 * namei for open - this is in fact almost the whole open-routine.
1281 * Note that the low bits of "flag" aren't the same as in the open
1282 * system call - they are 00 - no permissions needed
1283 * 01 - read permission needed
1284 * 10 - write permission needed
1285 * 11 - read/write permissions needed
1286 * which is a lot more logical, and also allows the "no perm" needed
1287 * for symlinks (where the permissions are checked later).
1290 int open_namei(const char * pathname, int flag, int mode, struct nameidata *nd)
1292 int acc_mode, error = 0;
1293 struct dentry *dentry;
1297 acc_mode = ACC_MODE(flag);
1299 /* Allow the LSM permission hook to distinguish append
1300 access from general write access. */
1301 if (flag & O_APPEND)
1302 acc_mode |= MAY_APPEND;
1304 /* Fill in the open() intent data */
1305 nd->intent.open.flags = flag;
1306 nd->intent.open.create_mode = mode;
1309 * The simplest case - just a plain lookup.
1311 if (!(flag & O_CREAT)) {
1312 error = path_lookup(pathname, lookup_flags(flag)|LOOKUP_OPEN, nd);
1319 * Create - we need to know the parent.
1321 error = path_lookup(pathname, LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE, nd);
1326 * We have the parent and last component. First of all, check
1327 * that we are not asked to creat(2) an obvious directory - that
1331 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1335 nd->flags &= ~LOOKUP_PARENT;
1336 down(&dir->d_inode->i_sem);
1337 dentry = __lookup_hash(&nd->last, nd->dentry, nd);
1340 error = PTR_ERR(dentry);
1341 if (IS_ERR(dentry)) {
1342 up(&dir->d_inode->i_sem);
1346 /* Negative dentry, just create the file */
1347 if (!dentry->d_inode) {
1348 if (!IS_POSIXACL(dir->d_inode))
1349 mode &= ~current->fs->umask;
1350 error = vfs_create(dir->d_inode, dentry, mode, nd);
1351 up(&dir->d_inode->i_sem);
1353 nd->dentry = dentry;
1356 /* Don't check for write permission, don't truncate */
1363 * It already exists.
1365 up(&dir->d_inode->i_sem);
1371 if (d_mountpoint(dentry)) {
1373 if (flag & O_NOFOLLOW)
1375 while (__follow_down(&nd->mnt,&dentry) && d_mountpoint(dentry));
1378 if (!dentry->d_inode)
1380 if (dentry->d_inode->i_op && dentry->d_inode->i_op->follow_link)
1384 nd->dentry = dentry;
1386 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode))
1389 error = may_open(nd, acc_mode, flag);
1402 if (flag & O_NOFOLLOW)
1405 * This is subtle. Instead of calling do_follow_link() we do the
1406 * thing by hands. The reason is that this way we have zero link_count
1407 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1408 * After that we have the parent and last component, i.e.
1409 * we are in the same situation as after the first path_walk().
1410 * Well, almost - if the last component is normal we get its copy
1411 * stored in nd->last.name and we will have to putname() it when we
1412 * are done. Procfs-like symlinks just set LAST_BIND.
1414 nd->flags |= LOOKUP_PARENT;
1415 error = security_inode_follow_link(dentry, nd);
1418 touch_atime(nd->mnt, dentry);
1419 nd_set_link(nd, NULL);
1420 error = dentry->d_inode->i_op->follow_link(dentry, nd);
1422 char *s = nd_get_link(nd);
1424 error = __vfs_follow_link(nd, s);
1425 if (dentry->d_inode->i_op->put_link)
1426 dentry->d_inode->i_op->put_link(dentry, nd);
1431 nd->flags &= ~LOOKUP_PARENT;
1432 if (nd->last_type == LAST_BIND) {
1433 dentry = nd->dentry;
1437 if (nd->last_type != LAST_NORM)
1439 if (nd->last.name[nd->last.len]) {
1440 putname(nd->last.name);
1445 putname(nd->last.name);
1449 down(&dir->d_inode->i_sem);
1450 dentry = __lookup_hash(&nd->last, nd->dentry, nd);
1451 putname(nd->last.name);
1456 * lookup_create - lookup a dentry, creating it if it doesn't exist
1457 * @nd: nameidata info
1458 * @is_dir: directory flag
1460 * Simple function to lookup and return a dentry and create it
1461 * if it doesn't exist. Is SMP-safe.
1463 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1465 struct dentry *dentry;
1467 down(&nd->dentry->d_inode->i_sem);
1468 dentry = ERR_PTR(-EEXIST);
1469 if (nd->last_type != LAST_NORM)
1471 nd->flags &= ~LOOKUP_PARENT;
1472 dentry = lookup_hash(&nd->last, nd->dentry);
1475 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1480 dentry = ERR_PTR(-ENOENT);
1485 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1487 int error = may_create(dir, dentry, NULL);
1492 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1495 if (!dir->i_op || !dir->i_op->mknod)
1498 error = security_inode_mknod(dir, dentry, mode, dev);
1503 error = dir->i_op->mknod(dir, dentry, mode, dev);
1505 inode_dir_notify(dir, DN_CREATE);
1506 security_inode_post_mknod(dir, dentry, mode, dev);
1511 asmlinkage long sys_mknod(const char __user * filename, int mode, unsigned dev)
1515 struct dentry * dentry;
1516 struct nameidata nd;
1520 tmp = getname(filename);
1522 return PTR_ERR(tmp);
1524 error = path_lookup(tmp, LOOKUP_PARENT, &nd);
1527 dentry = lookup_create(&nd, 0);
1528 error = PTR_ERR(dentry);
1530 if (!IS_POSIXACL(nd.dentry->d_inode))
1531 mode &= ~current->fs->umask;
1532 if (!IS_ERR(dentry)) {
1533 switch (mode & S_IFMT) {
1534 case 0: case S_IFREG:
1535 error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1537 case S_IFCHR: case S_IFBLK:
1538 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,
1539 new_decode_dev(dev));
1541 case S_IFIFO: case S_IFSOCK:
1542 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,0);
1552 up(&nd.dentry->d_inode->i_sem);
1560 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1562 int error = may_create(dir, dentry, NULL);
1567 if (!dir->i_op || !dir->i_op->mkdir)
1570 mode &= (S_IRWXUGO|S_ISVTX);
1571 error = security_inode_mkdir(dir, dentry, mode);
1576 error = dir->i_op->mkdir(dir, dentry, mode);
1578 inode_dir_notify(dir, DN_CREATE);
1579 security_inode_post_mkdir(dir,dentry, mode);
1584 asmlinkage long sys_mkdir(const char __user * pathname, int mode)
1589 tmp = getname(pathname);
1590 error = PTR_ERR(tmp);
1592 struct dentry *dentry;
1593 struct nameidata nd;
1595 error = path_lookup(tmp, LOOKUP_PARENT, &nd);
1598 dentry = lookup_create(&nd, 1);
1599 error = PTR_ERR(dentry);
1600 if (!IS_ERR(dentry)) {
1601 if (!IS_POSIXACL(nd.dentry->d_inode))
1602 mode &= ~current->fs->umask;
1603 error = vfs_mkdir(nd.dentry->d_inode, dentry, mode);
1606 up(&nd.dentry->d_inode->i_sem);
1616 * We try to drop the dentry early: we should have
1617 * a usage count of 2 if we're the only user of this
1618 * dentry, and if that is true (possibly after pruning
1619 * the dcache), then we drop the dentry now.
1621 * A low-level filesystem can, if it choses, legally
1624 * if (!d_unhashed(dentry))
1627 * if it cannot handle the case of removing a directory
1628 * that is still in use by something else..
1630 static void d_unhash(struct dentry *dentry)
1633 spin_lock(&dcache_lock);
1634 switch (atomic_read(&dentry->d_count)) {
1636 spin_unlock(&dcache_lock);
1637 shrink_dcache_parent(dentry);
1638 spin_lock(&dcache_lock);
1639 if (atomic_read(&dentry->d_count) != 2)
1644 spin_unlock(&dcache_lock);
1647 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
1649 int error = may_delete(dir, dentry, 1);
1654 if (!dir->i_op || !dir->i_op->rmdir)
1659 down(&dentry->d_inode->i_sem);
1661 if (d_mountpoint(dentry))
1664 error = security_inode_rmdir(dir, dentry);
1666 error = dir->i_op->rmdir(dir, dentry);
1668 dentry->d_inode->i_flags |= S_DEAD;
1671 up(&dentry->d_inode->i_sem);
1673 inode_dir_notify(dir, DN_DELETE);
1681 asmlinkage long sys_rmdir(const char __user * pathname)
1685 struct dentry *dentry;
1686 struct nameidata nd;
1688 name = getname(pathname);
1690 return PTR_ERR(name);
1692 error = path_lookup(name, LOOKUP_PARENT, &nd);
1696 switch(nd.last_type) {
1707 down(&nd.dentry->d_inode->i_sem);
1708 dentry = lookup_hash(&nd.last, nd.dentry);
1709 error = PTR_ERR(dentry);
1710 if (!IS_ERR(dentry)) {
1711 error = vfs_rmdir(nd.dentry->d_inode, dentry);
1714 up(&nd.dentry->d_inode->i_sem);
1722 int vfs_unlink(struct inode *dir, struct dentry *dentry)
1724 int error = may_delete(dir, dentry, 0);
1729 if (!dir->i_op || !dir->i_op->unlink)
1734 down(&dentry->d_inode->i_sem);
1735 if (d_mountpoint(dentry))
1738 error = security_inode_unlink(dir, dentry);
1740 error = dir->i_op->unlink(dir, dentry);
1742 up(&dentry->d_inode->i_sem);
1744 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
1745 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
1747 inode_dir_notify(dir, DN_DELETE);
1753 * Make sure that the actual truncation of the file will occur outside its
1754 * directory's i_sem. Truncate can take a long time if there is a lot of
1755 * writeout happening, and we don't want to prevent access to the directory
1756 * while waiting on the I/O.
1758 asmlinkage long sys_unlink(const char __user * pathname)
1762 struct dentry *dentry;
1763 struct nameidata nd;
1764 struct inode *inode = NULL;
1766 name = getname(pathname);
1768 return PTR_ERR(name);
1770 error = path_lookup(name, LOOKUP_PARENT, &nd);
1774 if (nd.last_type != LAST_NORM)
1776 down(&nd.dentry->d_inode->i_sem);
1777 dentry = lookup_hash(&nd.last, nd.dentry);
1778 error = PTR_ERR(dentry);
1779 if (!IS_ERR(dentry)) {
1780 /* Why not before? Because we want correct error value */
1781 if (nd.last.name[nd.last.len])
1783 inode = dentry->d_inode;
1785 atomic_inc(&inode->i_count);
1786 error = vfs_unlink(nd.dentry->d_inode, dentry);
1790 up(&nd.dentry->d_inode->i_sem);
1797 iput(inode); /* truncate the inode here */
1801 error = !dentry->d_inode ? -ENOENT :
1802 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
1806 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname, int mode)
1808 int error = may_create(dir, dentry, NULL);
1813 if (!dir->i_op || !dir->i_op->symlink)
1816 error = security_inode_symlink(dir, dentry, oldname);
1821 error = dir->i_op->symlink(dir, dentry, oldname);
1823 inode_dir_notify(dir, DN_CREATE);
1824 security_inode_post_symlink(dir, dentry, oldname);
1829 asmlinkage long sys_symlink(const char __user * oldname, const char __user * newname)
1835 from = getname(oldname);
1837 return PTR_ERR(from);
1838 to = getname(newname);
1839 error = PTR_ERR(to);
1841 struct dentry *dentry;
1842 struct nameidata nd;
1844 error = path_lookup(to, LOOKUP_PARENT, &nd);
1847 dentry = lookup_create(&nd, 0);
1848 error = PTR_ERR(dentry);
1849 if (!IS_ERR(dentry)) {
1850 error = vfs_symlink(nd.dentry->d_inode, dentry, from, S_IALLUGO);
1853 up(&nd.dentry->d_inode->i_sem);
1862 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
1864 struct inode *inode = old_dentry->d_inode;
1870 error = may_create(dir, new_dentry, NULL);
1874 if (dir->i_sb != inode->i_sb)
1878 * A link to an append-only or immutable file cannot be created.
1880 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1882 if (!dir->i_op || !dir->i_op->link)
1884 if (S_ISDIR(old_dentry->d_inode->i_mode))
1887 error = security_inode_link(old_dentry, dir, new_dentry);
1891 down(&old_dentry->d_inode->i_sem);
1893 error = dir->i_op->link(old_dentry, dir, new_dentry);
1894 up(&old_dentry->d_inode->i_sem);
1896 inode_dir_notify(dir, DN_CREATE);
1897 security_inode_post_link(old_dentry, dir, new_dentry);
1903 * Hardlinks are often used in delicate situations. We avoid
1904 * security-related surprises by not following symlinks on the
1907 * We don't follow them on the oldname either to be compatible
1908 * with linux 2.0, and to avoid hard-linking to directories
1909 * and other special files. --ADM
1911 asmlinkage long sys_link(const char __user * oldname, const char __user * newname)
1913 struct dentry *new_dentry;
1914 struct nameidata nd, old_nd;
1918 to = getname(newname);
1922 error = __user_walk(oldname, 0, &old_nd);
1925 error = path_lookup(to, LOOKUP_PARENT, &nd);
1929 if (old_nd.mnt != nd.mnt)
1931 new_dentry = lookup_create(&nd, 0);
1932 error = PTR_ERR(new_dentry);
1933 if (!IS_ERR(new_dentry)) {
1934 error = vfs_link(old_nd.dentry, nd.dentry->d_inode, new_dentry);
1937 up(&nd.dentry->d_inode->i_sem);
1941 path_release(&old_nd);
1949 * The worst of all namespace operations - renaming directory. "Perverted"
1950 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
1952 * a) we can get into loop creation. Check is done in is_subdir().
1953 * b) race potential - two innocent renames can create a loop together.
1954 * That's where 4.4 screws up. Current fix: serialization on
1955 * sb->s_vfs_rename_sem. We might be more accurate, but that's another
1957 * c) we have to lock _three_ objects - parents and victim (if it exists).
1958 * And that - after we got ->i_sem on parents (until then we don't know
1959 * whether the target exists). Solution: try to be smart with locking
1960 * order for inodes. We rely on the fact that tree topology may change
1961 * only under ->s_vfs_rename_sem _and_ that parent of the object we
1962 * move will be locked. Thus we can rank directories by the tree
1963 * (ancestors first) and rank all non-directories after them.
1964 * That works since everybody except rename does "lock parent, lookup,
1965 * lock child" and rename is under ->s_vfs_rename_sem.
1966 * HOWEVER, it relies on the assumption that any object with ->lookup()
1967 * has no more than 1 dentry. If "hybrid" objects will ever appear,
1968 * we'd better make sure that there's no link(2) for them.
1969 * d) some filesystems don't support opened-but-unlinked directories,
1970 * either because of layout or because they are not ready to deal with
1971 * all cases correctly. The latter will be fixed (taking this sort of
1972 * stuff into VFS), but the former is not going away. Solution: the same
1973 * trick as in rmdir().
1974 * e) conversion from fhandle to dentry may come in the wrong moment - when
1975 * we are removing the target. Solution: we will have to grab ->i_sem
1976 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
1977 * ->i_sem on parents, which works but leads to some truely excessive
1980 int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
1981 struct inode *new_dir, struct dentry *new_dentry)
1984 struct inode *target;
1987 * If we are going to change the parent - check write permissions,
1988 * we'll need to flip '..'.
1990 if (new_dir != old_dir) {
1991 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
1996 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2000 target = new_dentry->d_inode;
2002 down(&target->i_sem);
2003 d_unhash(new_dentry);
2005 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2008 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2011 target->i_flags |= S_DEAD;
2013 if (d_unhashed(new_dentry))
2014 d_rehash(new_dentry);
2018 d_move(old_dentry,new_dentry);
2019 security_inode_post_rename(old_dir, old_dentry,
2020 new_dir, new_dentry);
2025 int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2026 struct inode *new_dir, struct dentry *new_dentry)
2028 struct inode *target;
2031 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2036 target = new_dentry->d_inode;
2038 down(&target->i_sem);
2039 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2042 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2044 /* The following d_move() should become unconditional */
2045 if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME))
2046 d_move(old_dentry, new_dentry);
2047 security_inode_post_rename(old_dir, old_dentry, new_dir, new_dentry);
2055 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2056 struct inode *new_dir, struct dentry *new_dentry)
2059 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2061 if (old_dentry->d_inode == new_dentry->d_inode)
2064 error = may_delete(old_dir, old_dentry, is_dir);
2068 if (!new_dentry->d_inode)
2069 error = may_create(new_dir, new_dentry, NULL);
2071 error = may_delete(new_dir, new_dentry, is_dir);
2075 if (!old_dir->i_op || !old_dir->i_op->rename)
2078 DQUOT_INIT(old_dir);
2079 DQUOT_INIT(new_dir);
2082 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2084 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2086 if (old_dir == new_dir)
2087 inode_dir_notify(old_dir, DN_RENAME);
2089 inode_dir_notify(old_dir, DN_DELETE);
2090 inode_dir_notify(new_dir, DN_CREATE);
2096 static inline int do_rename(const char * oldname, const char * newname)
2099 struct dentry * old_dir, * new_dir;
2100 struct dentry * old_dentry, *new_dentry;
2101 struct dentry * trap;
2102 struct nameidata oldnd, newnd;
2104 error = path_lookup(oldname, LOOKUP_PARENT, &oldnd);
2108 error = path_lookup(newname, LOOKUP_PARENT, &newnd);
2113 if (oldnd.mnt != newnd.mnt)
2116 old_dir = oldnd.dentry;
2118 if (oldnd.last_type != LAST_NORM)
2121 new_dir = newnd.dentry;
2122 if (newnd.last_type != LAST_NORM)
2125 trap = lock_rename(new_dir, old_dir);
2127 old_dentry = lookup_hash(&oldnd.last, old_dir);
2128 error = PTR_ERR(old_dentry);
2129 if (IS_ERR(old_dentry))
2131 /* source must exist */
2133 if (!old_dentry->d_inode)
2135 /* unless the source is a directory trailing slashes give -ENOTDIR */
2136 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2138 if (oldnd.last.name[oldnd.last.len])
2140 if (newnd.last.name[newnd.last.len])
2143 /* source should not be ancestor of target */
2145 if (old_dentry == trap)
2147 new_dentry = lookup_hash(&newnd.last, new_dir);
2148 error = PTR_ERR(new_dentry);
2149 if (IS_ERR(new_dentry))
2151 /* target should not be an ancestor of source */
2153 if (new_dentry == trap)
2156 error = vfs_rename(old_dir->d_inode, old_dentry,
2157 new_dir->d_inode, new_dentry);
2163 unlock_rename(new_dir, old_dir);
2165 path_release(&newnd);
2167 path_release(&oldnd);
2172 asmlinkage long sys_rename(const char __user * oldname, const char __user * newname)
2178 from = getname(oldname);
2180 return PTR_ERR(from);
2181 to = getname(newname);
2182 error = PTR_ERR(to);
2184 error = do_rename(from,to);
2191 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2195 len = PTR_ERR(link);
2200 if (len > (unsigned) buflen)
2202 if (copy_to_user(buffer, link, len))
2209 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2210 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2211 * using) it for any given inode is up to filesystem.
2213 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2215 struct nameidata nd;
2216 int res = dentry->d_inode->i_op->follow_link(dentry, &nd);
2218 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2219 if (dentry->d_inode->i_op->put_link)
2220 dentry->d_inode->i_op->put_link(dentry, &nd);
2226 __vfs_follow_link(struct nameidata *nd, const char *link)
2235 if (!walk_init_root(link, nd))
2236 /* weird __emul_prefix() stuff did it */
2239 res = link_path_walk(link, nd);
2241 if (current->link_count || res || nd->last_type!=LAST_NORM)
2244 * If it is an iterative symlinks resolution in open_namei() we
2245 * have to copy the last component. And all that crap because of
2246 * bloody create() on broken symlinks. Furrfu...
2249 if (unlikely(!name)) {
2253 strcpy(name, nd->last.name);
2254 nd->last.name = name;
2258 return PTR_ERR(link);
2261 int vfs_follow_link(struct nameidata *nd, const char *link)
2263 return __vfs_follow_link(nd, link);
2266 /* get the link contents into pagecache */
2267 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2270 struct address_space *mapping = dentry->d_inode->i_mapping;
2271 page = read_cache_page(mapping, 0, (filler_t *)mapping->a_ops->readpage,
2275 wait_on_page_locked(page);
2276 if (!PageUptodate(page))
2282 page_cache_release(page);
2283 return ERR_PTR(-EIO);
2289 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2291 struct page *page = NULL;
2292 char *s = page_getlink(dentry, &page);
2293 int res = vfs_readlink(dentry,buffer,buflen,s);
2296 page_cache_release(page);
2301 int page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2304 char *s = page_getlink(dentry, &page);
2312 void page_put_link(struct dentry *dentry, struct nameidata *nd)
2314 if (!IS_ERR(nd_get_link(nd))) {
2316 page = find_get_page(dentry->d_inode->i_mapping, 0);
2320 page_cache_release(page);
2321 page_cache_release(page);
2325 int page_follow_link(struct dentry *dentry, struct nameidata *nd)
2327 struct page *page = NULL;
2328 char *s = page_getlink(dentry, &page);
2329 int res = __vfs_follow_link(nd, s);
2332 page_cache_release(page);
2337 int page_symlink(struct inode *inode, const char *symname, int len)
2339 struct address_space *mapping = inode->i_mapping;
2340 struct page *page = grab_cache_page(mapping, 0);
2346 err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
2349 kaddr = kmap_atomic(page, KM_USER0);
2350 memcpy(kaddr, symname, len-1);
2351 kunmap_atomic(kaddr, KM_USER0);
2352 mapping->a_ops->commit_write(NULL, page, 0, len-1);
2354 * Notice that we are _not_ going to block here - end of page is
2355 * unmapped, so this will only try to map the rest of page, see
2356 * that it is unmapped (typically even will not look into inode -
2357 * ->i_size will be enough for everything) and zero it out.
2358 * OTOH it's obviously correct and should make the page up-to-date.
2360 if (!PageUptodate(page)) {
2361 err = mapping->a_ops->readpage(NULL, page);
2362 wait_on_page_locked(page);
2366 page_cache_release(page);
2369 mark_inode_dirty(inode);
2373 page_cache_release(page);
2378 struct inode_operations page_symlink_inode_operations = {
2379 .readlink = generic_readlink,
2380 .follow_link = page_follow_link_light,
2381 .put_link = page_put_link,
2384 EXPORT_SYMBOL(__user_walk);
2385 EXPORT_SYMBOL(follow_down);
2386 EXPORT_SYMBOL(follow_up);
2387 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2388 EXPORT_SYMBOL(getname);
2389 EXPORT_SYMBOL(lock_rename);
2390 EXPORT_SYMBOL(lookup_create);
2391 EXPORT_SYMBOL(lookup_hash);
2392 EXPORT_SYMBOL(lookup_one_len);
2393 EXPORT_SYMBOL(page_follow_link);
2394 EXPORT_SYMBOL(page_follow_link_light);
2395 EXPORT_SYMBOL(page_put_link);
2396 EXPORT_SYMBOL(page_readlink);
2397 EXPORT_SYMBOL(page_symlink);
2398 EXPORT_SYMBOL(page_symlink_inode_operations);
2399 EXPORT_SYMBOL(path_lookup);
2400 EXPORT_SYMBOL(path_release);
2401 EXPORT_SYMBOL(path_walk);
2402 EXPORT_SYMBOL(permission);
2403 EXPORT_SYMBOL(unlock_rename);
2404 EXPORT_SYMBOL(vfs_create);
2405 EXPORT_SYMBOL(vfs_follow_link);
2406 EXPORT_SYMBOL(vfs_link);
2407 EXPORT_SYMBOL(vfs_mkdir);
2408 EXPORT_SYMBOL(vfs_mknod);
2409 EXPORT_SYMBOL(vfs_permission);
2410 EXPORT_SYMBOL(vfs_readlink);
2411 EXPORT_SYMBOL(vfs_rename);
2412 EXPORT_SYMBOL(vfs_rmdir);
2413 EXPORT_SYMBOL(vfs_symlink);
2414 EXPORT_SYMBOL(vfs_unlink);
2415 EXPORT_SYMBOL(generic_readlink);